Download presentation
Presentation is loading. Please wait.
1
Near-infrared (NIR) Single Photon Counting Detectors (SPADs)
Chong Hu, Minggou Liu, Joe C. Campbell & Archie Holmes ECE Department University of Virginia
2
Outline Introduction to Single Photon Counting Detectors (SPADs)
Current States of near-infrared SPADs Summary and Future Goals
3
APD PMT Superconductors Hotspot Generation and Resistive Barrier
Mature Novel Superconductors Hotspot Generation and Resistive Barrier High efficiency Low dark counts No afterpulsing T < 1K!! APD t Electron & Hole avalanche multiplication Good efficiency Acceptable dark counts Afterpulsing PMT High gain Low dark current Low noise Low quantum efficiency Large, bulky Expensive High voltage Fragile Ambient light catastrophic 2000 V First let say a little about some of the candidates. Obviously, we have to start with the reigning champion photomultipliers. High gain, low noise, and acceptable responsivity in the visible. Avalanche photodiodes, particularly Geiger mode devices, have made tremendous strides in the last two decades. Good quantum efficiency, moderate noise, but still unacceptable dark count rates (false positives) and afterpulsing seriously restricts sampling rates. Some of the superconducting devices show a lot of promise. OPERATION… However, I don’t think they will be viable because of the impractical cooling that is required. The same restriction applies to current proposals for quantum dot single photon counting devices.
4
Geiger-mode operation
Geiger mode - APD functions as a switch Analog Digital Responsivity Single photon detection efficiency Dark current Probability of dark count Concept of excess noise does not apply! Geiger-mode operation Single photon input time APD output Discriminator level time time Digital comparator output time time Successful single photon detection Photon absorbed but insufficient gain – missed count Dark count – from dark current
5
Current Current Current Current Voltage Voltage
Linear Geiger on mode mode on avalanche Linear Geiger quench mode mode Current Current Current Current off off I think I have already covered these … read arm Vdc + DV V V br br Voltage Voltage
6
Performance Parameters
Single photon input Photon detection efficiency (PDE) The probability that a single incident photon initiates a current pulse that registers in a digital counter Dark count Rate (DCR)/Probability (DCP) The probability that a count is triggered by dark current instead of incident photons time APD output Discriminator level I think I have already covered these … read time time Digital comparator output time time Successful single photon detection Photon absorbed but insufficient gain – missed count Dark count – from dark current
7
Photon Detection Efficiency
hPDE = hexternal x hcollection x Pavalanche For Geiger-mode APDs, commonly referred to as single photon avalanche detectors, or SPADs, the single photon detection efficiency is the product of the external quantum efficiency and the avalanche probability. Put simply, first you have to create an electron-hole pair by absorption of a photon and then one of those carriers has to initiate an avalanche breakdown. As you can see, the avalanche breakdown probability increases with excess bias above breakdown. InGaAsP , 1.3 mm InGaAs mm 7
8
40m-diameter In0.53Ga0.47As/InP
Dark current 0.18nA at 95% of Vbr at 297K 0.15pA at 95% of Vbr at 200K Good enough?
9
Photon Detection Efficiency
1.06 μm SPADs: DCR vs. PDE InGaAsP absorber lower generation-recombination dark current DCR approaching Si SPAD DCR with greatly increased PDE Si SPADs have PDE < 2% at 1.06 μm 105 300K 273K – 295K * 104 259 K 250 K Dark Count Rate (Hz) 103 237 K 80 μm dia. InGaAsP SPAD 1-ns gated operation 500 kHz repetition rate 0.1 photon/pulse AP probability < 10-4 per gate 102 101 0% 10% 20% 30% 40% 50% Photon Detection Efficiency
10
Dark Count Probability versus Photon Detection Efficiency
10 -2 InGaAs/InP, 300K 1550 nm (2007) 10 -3 InGaAs/InP (2007) 10 -4 Dark Count Probability 10 -5 240K We did not push hard on dark count probability this year because Ray suggested other thrust areas for us. However, we have established a strong collaboration with Princeton Lightwave in developing SPADs – exchanged data, visits, wild ideas, and they served on the qualifier and PhD final committees of our students. This shows a recent result where the dark count probability has been reduced by over an order of magnitude – at 243K approaching that of the best room temperature Si devices. You may wonder about the hockey stick shape of these curves. They were measured with the IBM, single photon counting in a box apparatus that can only supply excess voltage of 4 V. Higher detection efficiencies are achieved by using higher excess bias. The only way that could be accomplished was to move the dc bias closer to breakdown which caused the dark counts to increase. We anticipate that soon we can straighten these out. Note that these results will also allow us to operate at much higher temperatures – hopefully with only modest cooling. 200K 10 -6 10 -7 5 10 15 20 25 30 35 40 45 50 Single Photon Detection Efficiency (%) 10
11
Current Current Current Current Voltage Voltage
Linear Geiger on mode mode on avalanche Linear Geiger quench mode mode Current Current Current Current off off I think I have already covered these … read arm Vdc + DV V V br br Voltage Voltage
12
Quenching Techniques Quenching circuit Passive Quenching
Quench avalanche current Reset the device Passive Quenching Quenched by discharging capacitance Slow recharge Active Quenching Raise the anode voltage Quick recharge Gated Quenching Vq>VEX Once an avalanche event has occurred, the APD will remain in the high current mode until the avalanche is quenched by lowering the bias voltage. Any quenching circuit has to accomplish two tasks. The first one is to quench the current by lowering the voltage on the device when an avalanche happens. The second one is to reset the device for subsequent detection by raising the voltage above breakdown. In passive quenching, when the APD turns on, the capacitance is discharged which reduces the bias voltage and quenches the avalanche. The capacitance is then recharged through the large series resistor. For active quenching a circuit senses the avalanche event and feeds back a reset pulse. I will discuss our recent work on this later. Finally, we have gated quenching, which is the technique we used when we began this program.
13
Afterpulsing Number of trapped carriers Initial avalanche
Biasing scheme Initial avalanche Released carriers from traps During an avalanche event, traps in the multiplication region capture carriers. As those carriers are released, they can trigger subsequent avalanches after the APD has been quenched and reset. This is illustrated in the graph on the right. After an avalanche event the bias is kept below breakdown for a period referred to as the hold off time. As you can see in the graph, if the hold off time is too short to allow all of the trapped carriers to escape, the dark count rate will become excessive due to this afterpulsing effect. Lower temperature where the trap lifetimes are longer, require longer hold off times. For optical communications, such as quantum cryptography, this limits the data transmission rate. For imaging, it limits the number of bits that we can collect in each pixel during a frame. So it is important to try to find ways to reduce the effect of afterpulsing.
14
Reduction of Afterpulsing: Decreasing Charge Flow
Passive quenching In the initial part of Phase I, we concentrated on pulsed quenching in order to study the device physics of the SPADs. However, as I have said, for pulsed quenching you have to bias the APD above synchronously with the arrival of the photon, which is fine for optical communications but not so appropriate for imaging where you don’t know the arrival time of the photons. That brings us back to active or passive quenching. Passive quenching is better for reducing afterpulsing because the time constants associated with the active feedback loop allow too much current to flow before quenching. For passive quenching, when the device turns on, the capacitor discharges until the voltage drops sufficiently to “turn the avalanche off” – illustrated in the top graph. When the device turns on, the capacitor discharges until the voltage drops sufficiently to “turn the avalanche off” – illustrated in the top graph. The capacitance then recharges through the large series resistance. However, there is a problem associated with the magnitude of the quenching resistor. If the resistor is very large we get the fastest quenching, which reduces the trapped charge and thus reduces afterpulsing, but it takes a very long time to recharge for the next photon. On the other hand, if the resistor is small, the recharge can occur faster but the voltage may increase before all the trapped carriers have been emitted, which will result in more after pulsing. The total charges flowing through device: Q=(Cs+Cd)Vex Cd: device capacitance Cs: stray capacitance
15
Passive Quenching with Active Reset (PQAR)
To deal with that we have developed a new quenching circuit, passive quenching with active reset, PQAR. It has a very large quenching resistor for rapid quenching. It also keeps the excess bias very low while the traps discharge. This is the passive quenching aspect of this circuit. As the avalanche is initiated, the negative voltage pulse at the cathode of the SPAD is coupled into a pulse generator which after a preset delay sends a pulse to the FET which shorts out the quenching resistor and allows rapid reset through the 50 W resistor, which is very fast because the 10 MW resistor has been effectively shorted or bypassed. The net result is that the SPAD is held in the off position to allow the trapped carriers to escape and then the voltage is quickly reset to the detection state. The switch is open during recharge to prevent transients from cause false counts.
16
PQAR at 230K Compare with gated mode results Vb 34s 15s
70 Vex=2.6V 60 Vex=2.0V 50 Vex=1.6V Hold-off = 15s 40 Measured counts x 1000 (/s) 30 20 10 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 CW laser power (fW) Voltage on device Compare with gated mode results 34s 15s Vb 10-4 10-5 10-6 Dark count probability (/ns) NEP ~ W/Hz1/2 This slide illustrates the benefit of our active quenching circuit. At 230K, the dark count rate corresponds to a dark count every 34 microseconds and we have a hold off time of 15 microseconds. The DCR is the number of dark counts divided by the time that it was biased above breakdown which is 1-number of dark counts x hold off time. Total count rate … similar…. Which is …. So the Photon Counting rate is photon flux x PDE. Upper left graph shows the total number of counts versus cw optical flux – as you can see it begins to saturate (approach the DCR) near a fW. The graph at the upper right shows the photon count rate versus the incident flux. The response is linear over several orders of magnitude and the intercept is the detection efficiency. The most important thing, however, is that here we show that we get about the same dark count probability and detection efficiency but we have improved the duty cycle by over two orders of magnitude.
17
Gated Quenching of a SPAD
AC pulse Excess bias width (V ) ex Total bias on APD V br V dc Time Laser pulse Time Avalanche This illustrates single photon detection with gated quenching. The bias is maintained below breakdown and then pulse biased above breakdown synchronized with the arrival of the photon. This is an excellent technique for studying the physics of Geiger-mode APDs and was the technique we used at the beginning of Phase I. As I will discuss later, we have developed a new quenching circuit that is more appropriate for imaging applications. In operation: If an incident photon results in an avalanche event, a count will be registered. However, sometimes the photon does not create an avalanche event – this is a false negative. The relative frequency of these two events determines the single photon detection efficiency. On the other hand, some times we get an avalanche event even though there was no incident photon. This is caused by the carriers in the dark current. This is a false positive. This will be expressed as the dark count rate. pulse Time Avalanche pulse Missed photon Avalanche pulse due to incident photon (false negative) due to dark carriers (false positive) ß Photon Detection Efficiency Dark Count Rate
18
Gated-PQAR Compared to PQAR Suppressed dark counts by gated bias
Cag Cac Vbias Rs=50Ω Amplifier A Counter Vgate Compared to PQAR Suppressed dark counts by gated bias Reduced complexity Array operation: recharge together, quench separately The transistor can be HBT monolithically integrated on the SPAD, and the its gate/base input can be shared over the whole array.
19
Photon Detection Efficiency (%)
Gated Quenching and Gated PQAR 10 8 6 4 Dark Count Rate (kHz) 220K 200K 180K 2 220K 200K 180K gated-quenching gated-PQAR 1 5 10 15 20 25 30 35 Photon Detection Efficiency (%)
20
Gated Quenching and Gated PQAR Dark Count Probability
220K, Vex = 5.6% 10 -3 200K, Vex = 6.0% 180K, Vex = 6.2% 220K, Vex = 5.6% 200K, Vex = 6.0% 180K, Vex = 6.2% Dark Count Probability gated-quenching 10 -4 gated-PQAR 10 -5 0.01 0.1 1 10 Repetition Rate (MHz)
21
Conclusions Performance of Geiger-mode APDs is improving rapidly
Acceptable detection efficiencies and dark count probability levels Getting a better control over the afterpulsing problem
22
Future Goals Move closer to quantum limited detection
Dark Current 0 Quantum Efficiency 100% Read Noise 0 Move to longer wavelengths Do photon number resolving
23
High QE Structure
24
Type-II Quantum Wells (QWs)
Formed between materials with staggered band line-ups Electrons and holes are confined in adjoining layers Spatially indirect absorption and emission Smaller effective bandgap for long-wavelength operation E GaAs0.5Sb0.5 0.79eV ΔEc=0.236eV CB 0.49eV ≈ 2.5 μm VB Ga0.47In0.53As 0.77eV ΔEv=0.247eV x
25
Where we are now (pin devices)
Rogalski, A., Progress in Quant. Elec., 27(2-3), pp 59 (2003) -2 V bias (200K) -2 V bias (RT)
26
Gated-PQAR for Synchronized Detection
Gated Quench Compared to gated quench Comparable circuit complexity Wider AC pulses: easier to generate and synchronize Uniform output pulse shape, good for photon-number-resolution with multiplexing photon photon photon AC output Gated-PQAR photon photon photon Vbias Transistor on: low resistance for fast reset Vgate Transistor off: high resistance for fast passive quench Vdiode output
27
Questions??
28
Simulated Breakdown Probabilities
J. P. R. David, University of Sheffield InP AlInAs Decreasing thickness: 0.2 m m, 0.5 m, 1.0 Previously, we have simulated the breakdown probability for different multiplication layers. Recently, our colleagues at the University of Sheffield have done a similar simulation and found results similar to ours. This is their simulation for InP and AlInAs of varying width. At present the devices we use are designated by the red curve. As you can see, we should be able to get better results with AlInAs and by decreasing the thickness of the multiplication layer. However, the multiplication layer can not be reduced too much or tunneling will lead to higher dark current and thus higher dark counts. While these curved may not appear to be greatly different, it would appear that we could pick up about 15% in single photon detection efficiency by going to a AlInAs multiplication layer that is less than 1 micron thick. We would like to work on this in the coming year. 28
29
20 x 20 Array – 98% yield 4 x 4 Subarray – uniform single-photon response
30
Afterpulsing Probability vs. Total Charge
AC pulse Delay 1ms Period Laser pulse This illustrates the double pulse technique where we trigger an avalanche event in the first pulse and then look for a dark count in the second pulse. What we find is that the after pulse probability depends only on the total charge in the first avalanche. It does not matter if we increase the pulse height or the pulse width, the only parameter that effects the afterpulsing is the total charge that passes through the device during an avalanche event.
31
Total Charge Flow
32
“Effective Excess Noise Factor”
Recently a number of papers in this area have started using something they refer to as the effective excess noise factor. Basically, it is a measure of the variation in pulse height to the counter for the avalanche events. I think it is a useful concept but an unfortunate, misleading term. Obviously, if we set the threshold level for the counter “here”, we don’t want any of the pulses anywhere near – and you can see that is the case. Regardless of my feelings about the term, I think our PQAR quenching circuit has achieved the best “effective excess noise factor” reported to date.
33
Pixel Level Monolithic Integration of Active Switching Elements
Reduced parasitics Faster quenching Reduced afterpulsing Increased transmission and sampling rates Packaging advantages R R =50 =50 W s s Counter Counter Active switching element V V bias bias 33
Similar presentations
